---
title: Vector Database & Sentiment Analysis Tool Demo (2026-02-10)
type: article
created: '2026-02-10'
updated: '2026-04-05'
source_docs:
- raw/2026-02-10-weekly-call-w-sebastian-121237915.md
tags:
- client-management
- tooling
- ai
- sentiment-analysis
- vector-database
- operations
layer: 2
client_source: null
industry_context: null
transferable: true
---

# Vector Database & Sentiment Analysis Tool Demo (2026-02-10)

## Overview

During the [[meetings/2026-02-10-weekly-call-sebastian|2026-02-10 weekly call with Sebastian]], Mark demoed an internal client intelligence tool built on a vector database. The tool ingests all internal communications across every platform and uses AI to surface sentiment analysis, risk flags, and activity trends — enabling proactive client management without manually reviewing hundreds of documents.

This tool represents a significant operational capability: account managers can get a real-time read on client health across their entire book of business in minutes rather than hours.

## How It Works

The system ingests and indexes communications from:

- **Email** (inbound, outbound, and internal)
- **Slack**
- **ClickUp tasks**
- **Google Drive** (Docs, Sheets, Presentations, PDFs — not images)
- **Meeting transcripts**

Documents are chunked and stored in a **vector database**, which enables semantic search — you can ask natural-language questions like "when did we last discuss X?" or "what did the client say about Y?" and the system retrieves relevant chunks across all sources.

## Key Features

### Sentiment Analysis & Risk Flags

For each client, the tool analyzes the last 30 days of communications and produces:

- **Overall sentiment** (e.g., positive and stable, negative, mixed)
- **Risk flags** — specific issues surfaced from the data, such as:
  - Timeline slippage / overdue deliverables
  - Tone trend changes (e.g., client becoming more terse or critical)
  - Longstanding blockers (e.g., unresolved technical issues)
  - Gaps in communication
- **Key relationship indicators** — characterizes the nature of the client relationship based on language patterns (e.g., "collaborative and appreciative," "transactional")

> **Example from the demo:** The tool flagged [[clients/overhead-door/_index|Overhead Door]] as having critical issues — Google Ads verification stalled for 12+ months, repeated design asset delays, 132 SEO warnings, and sporadic blog execution — despite an overall positive sentiment score. This surfaced actionable blockers that weren't visible from meeting notes alone.

### Activity Dashboard

Each client has a communication activity view showing:

- Total document count by type (emails, meetings, Slacks, ClickUp tasks, Drive docs)
- Inbound vs. outbound vs. internal email breakdown
- Activity trend over time (volume of communication by week/month)

Low activity periods are immediately visible — useful for spotting clients who have gone quiet, which can be an early churn signal.

### Client Stream View

A chronological feed of all communications for a given client. From this view you can:

- Request an AI-generated summary of the last N days of activity
- Open any individual document (email, meeting, task) directly
- Download or copy summaries in Markdown format for use in reports or Google Docs

### Natural Language Search (In Development)

A "search and ask" interface is being built to allow freeform queries across all client data — e.g., "What did Overhead Door say about their competitor last month?" or "When did we last send a report to Aviary?"

## Operational Value

| Use Case | How the Tool Helps |
|---|---|
| Weekly client prep | Pull a 7-day summary before each meeting instead of re-reading email threads |
| Churn risk detection | Sentiment flags surface dissatisfaction before the client raises it directly |
| Account handoffs | New AM can get up to speed on relationship history quickly |
| Reporting | Copy Markdown summaries directly into Google Docs for client-facing reports |
| Accountability | Overdue deliverables and blockers are surfaced automatically from ClickUp + email |

## Practical Notes

- Summaries are output in **Markdown** and can be pasted directly into Google Docs via Edit → Paste from Markdown
- The tool currently covers ~250+ documents per client across all sources
- Vector databases work by chunking documents and retrieving semantically relevant chunks based on your query — not keyword matching
- The system is built and maintained by Mark; additional features (ask/search interface) are actively in development as of February 2026

## Related

- [[meetings/2026-02-10-weekly-call-sebastian|Weekly Call w/ Sebastian — Team Restructuring & Hiring Kickoff (2026-02-10)]]
- [[clients/overhead-door/_index|Overhead Door]] — first client flagged by the sentiment tool with critical issues
- [[knowledge/client-management/proactive-client-management|Proactive Client Management]]
- [[knowledge/operations/ai-tooling|AI Tooling & Automation]]